convolution#

datacheese.convolution.conv2d(img, kernel, stride=(1, 1), padding=0, fill=0)#

Perform 2D convolution operation over image array using kernel.

Parameters:

  • img (numpy.ndarray) – 2D image array to perform convolution on.

  • kernel (numpy.ndarray) – 2D kernel array.

  • stride (tuple) – 2-value tuple representing stride in each dimension.

  • padding (int) – Number of layers of padding to add around image.

  • fill (float) – Fill value to use when padding.

Returns:

out – 2D output array.

Return type:

numpy.nadarray

Examples

>>> import numpy as np
>>> from datacheese.convolution import conv2d

Define image and kernel:

>>> img = np.array(
...     [
...         [5, -2, 8, 1],
...         [3, -1, 0, -2],
...         [-9, 2, 8, -3],
...         [4, 7, -3, 4],
...     ]
... )
>>> kernel = np.array(
...     [
...         [-2, 3],
...         [0, 1],
...     ]
... )

Perform convolution of kernel over image using strides of 2 in both dimensions:

>>> conv2d(img, kernel, stride=(2, 2))
array([[-17, -15],
       [ 31, -21]])